Proceedings of the 1st Joint SLTU and CCURL Workshop (SLTU-CCURL 2020), pages 88–96 Language Resources and Evaluation Conference (LREC 2020), Marseille, 11–16 May 2020 c European Language Resources Association (ELRA), licensed under CC-BY-NC
Design and Evaluation of a Smartphone Keyboard for Plains Cree Syllabics
Eddie Antonio Santos∗†, Atticus G. Harrigan† ∗National Research Council Canada [email protected] †Alberta Language Technology Lab University of Alberta Edmonton, Alberta, Canada {easantos,Atticus.Harrigan}@ualberta.ca Abstract Plains Cree is a less-resourced language in Canada. To promote its usage online, we describe previous keyboard layouts for typing Plains Cree syllabics on smartphones. We describe our own solution whose development was guided by ergonomics research and corpus statistics. We then describe a case study in which three participants used a previous layout and our own, and we collected quantitative and qualitative data. We conclude that, despite observing accuracy improvements in user testing, introducing a brand new paradigm for typing Plains Cree syllabics may not be ideal for the community.
Keywords: cree, syllabics, smartphone
1. Introduction family. It is the westernmost language of the Cree lan- guage continuum, and has been historically spoken by the nêhiyawak people in an area spanning present day north and central Alberta, central Saskatchewan, and the eastern edge of British Columbia. As with many Indigenous languages in Canada, the use of Plains Cree was actively suppressed by government and institutional efforts (The Truth and Rec- onciliation Commission of Canada, 2015). Plains Cree is written primarily in two orthographies. The 123 (Cree Syllabics) ᓀᐦᐃᔭᐍᐏᐣ most commonly supported orthography by computers and smartphones is the standard Roman orthography (SRO), Figure 1: The virtual keyboard layout introduced in this pa- an alphabet that borrows the consonants ptkcshmnywrl and per for typing Plains Cree syllabics on iOS and Android vowels êiao from the Latin script (ISO 15924: Latn), and smartphones. denotes long vowels with either macron (ē ī ō ā) or circum- flex (ê î ô â) diacritics. Note that there is no short e vowel. To reclaim an minority language in the digital space, the language must be used and shared online. However, mi- nority language users of new media often resort to a ma- 3. ᒐᐦᑭᐯᐦᐃᑲᓇ: Writing in Syllabics jority language when writing online (Keegan et al., 2015). Syllabics or ᒐᐦᑭᐯᐦᐃᑲᓇ (cahkipêhikana) (ISO 15924: Ca- There is a need for quality mobile support for minority lan- ns) is a writing system created in the 1800s to write the guages (Cassels, 2019), such as Plains Cree. Having a good Swampy Cree language (Stevenson, 1999). Syllabics are mobile keyboard enables the language to be used online, used to write many Canadian Indigenous languages and of- and encourages the language’s use at home and in the class- ten have different conventions language to language. Such room. There is no well-established keyboard layout for languages include Inuktut, Eastern Cree dialects, and Dene Plain Cree syllabics, although there have been plenty of al- languages, among others. Their use for Plains Cree will be ternatives for smartphones (First Peoples’ Cultural Council, the focus of this paper. 2016; van Esch et al., 2019; Houle, 2018; Moshagen et al., The syllabics writing system is similar to an abugida—its 2016). In this paper, we explore the existing keyboard lay- primary graphemes represent a consonant-vowel pair, with outs, describe a few areas of improvement, and propose our the shape of each grapheme indicating the consonant of the own smartphone keyboard layout for writing in Plains Cree pair, and the orientation of the grapheme indicating the syllabics (Figure 1). We describe how we created our layout vowel of the pair. basing its key locations on ergonomics research and corpus For example, the graphemes ᐯ, ᐱ, ᐳ, ᐸ all have the statistics. We then describe a case study wherein we had same shape, which represents a syllable starting with the participants use our keyboard and compared it to an exist- /p/ sound; these graphemes represent the syllables /pe/, /pɪ/, ing solution. We conclude by discussing the preferences we /po/, /pɐ/, respectively. The graphemes ᑲ, ᓴ, ᒐ, ᒪ are all learned from the case study. oriented in the same way (the tips “point” to the southwest) which means they all contain the /ɐ/ vowel; these graphemes 2. nêhiyawêwin: the Cree Language represent the syllables /kɐ/, /sɐ/, /t͡ sɐ/, /mɐ/, respectively. Plains Cree (ISO 639-3: crk) or its endonym, ᓀᐦᐃᔭᐍᐏᐣ Syllables in Plains Cree have the general shape (C)(w)V- (nêhiyawêwin) is a member of the Algonquian language (C)(C), where C is a consonant and V is a vowel, and w is
88 Example 4.1 The FirstVoices Layout Category Elaboration Syllabic IPA SRO V Vowel ᐋ /ɑː/ â wV Syllable with /w/ as ᐚ /wɑː/ wâ onset CV Syllable with con- ᒌ /t͡ siː/ cî sonant onset CwV Syllable with con- ᓎ /nwɑː/ nwâ sonant + /w/ onset final Coda ᐤ /w/ w
Table 1: The different syllabic categories used in this paper. Figure 2: The FirstVoices Plains Cree syllabics layout.
The ᓀᐦᐃᔭᐍᐏᐣ (Plains Cree) keyboard layout (Figure 2) the /w/ sound1 (Wolfart, 1996). Given that most syllables was created by the First Peoples’ Cultural Council as part of in Plains Cree are predominately CV or CwV, the syllabics their FirstVoices Keyboards app (2016). The FirstVoices writing system is particularly well-suited for writing Plains Keyboards app is a suite of smartphone keyboard layouts Cree. Further, the writing system is thought to be more In- that covers all Indigenous languages in Canada. The tech- digenous and authentic than using the borrowed Latin writ- nology that allows the creation of many keyboard layouts ing system, thus it holds greater cultural value. is Keyman (2020), a keyboard creation engine. Keyman 3.1 Diacritics is embedded within the FirstVoices Keyboards app, and all of the keyboard layouts it provides—including The onset of a syllable, if present, is a consonant, or a conso- ᓀᐦᐃᔭᐍᐏᐣ (Plains Cree)—can be used standalone with the (separate) nant with an intervening /w/ sound. The /w/ sound is written Keyman app. Throughout the rest of this paper, we will re- in syllabics as a dot to the right-side of the (C)V syllabic. fer to this layout as the FirstVoices layout. For example, ᑌ represents the /te/ syllable; ᑘ, with a dot to the right, represents the /twe/ syllable. The FirstVoices layout uses a two page system: the initial Long vowels are denoted by putting a dot on top of a syl- page (Figure 2) displays all V and CV syllabics that have labic; for example, given the short vowels ᐃ (/ɪ/), ᐅ (/o/), the underlying /ɐ/ and /ɪ/ vowel. Below the CV syllabics ᐊ (/ɐ/), adding a dot diacritic on top of these vowels yields are all of the finals, with the exception of ᕽ. Pressing the the long vowels, ᐄ (/iː/) ᐆ (/oː/), and ᐋ (/ɑː/), respectively. key switches to the second page, which displays all V Note: although ᐁ (/e/) is considered to be a long vowel, it and CV syllabics that have the underlying /e/ and /o/ vowel, does not take a diacritic, as there is no short /e/ vowel. as well as keys for ᓬ (/l/), ᕒ (/r/), and ᕽ. To type a syllabic with a long vowel, the typist first types 3.2 Finals the V or CV syllabic with the corresponding short vowel. Codas in Plains Cree, if present, are one or two consonants Then, the typist must press the ̇ key, which has the effect written using syllabics called finals. Finals appear any- of adding a long vowel dot to the previously entered V or where in a word where a consonant cannot be paired with CV syllabic. To type a wV or CwV syllabic, one must first a vowel or an intervening /w/. For example, the word for type the corresponding V or CV syllabic, including long horse, /mɪstɐtɪm/, is written as ᒥᐢᑕᑎᒼ in syllabics. The fi- vowel dot, and then press the ᐧ key; this places a middle nals correspond to /s/ and /m/ phonemes, respectively, and dot after the corresponding syllabic. are the two consonants in the word that are not followed by a This layout has the advantage of having a shallow learning vowel or a /w/ + vowel pair. There are three special finals: ᐦ curve; All V and CV syllabics are one or two presses away. for writing /h/ anywhere in a word; ᐤ for words that end with However, since there are 40 V and CV syllabics, plus 13 fi- /w/; and ᕽ for words that end with /hk/; Every consonant in nals, this keyboard makes the decision to split them in half. Plains Cree has a corresponding final syllabic, though these Notably, syllables with the vowels /ɐ/ or /ɪ/ are more com- are not transparently related to their corresponding CV syl- mon in Plains Cree, so the keyboard made the right choice labic shape. in prioritizing which syllabics are immediately selectable. That said, the keyboard is still cramped; on our test device, 3.3 Categorizing Syllabics most keys on this layout are 5mm wide. In this paper, we define five categories for the Plains Cree The FirstVoices layout produces dubious Unicode charac- syllabics: V, wV, CV, CwV, and final (Table 1). The key- ter output. The Unified Canadian Aboriginal Syllabics boards described in this paper (Sections 4, 5) each have dif- block of Unicode defines unique characters for all V, wV, ferent methods of typing characters in each category. CV, and CwV syllabics and includes finals specifically for use in Western Cree syllabics (The Unicode Consor- 4. Prior Work tium, 2019b). The FirstVoices layout does not produce pre- We describe an existing solution to type Plains Cree syllab- composed characters for CwV syllabics; instead, it appends ics on smartphones: the FirstVoices Keyboards app. U+1427 CANADIAN SYLLABICS FINAL MIDDLE DOT to em- ulate a /w/ diacritic, despite the character actually encoding 1This is an oversimplification of Plains Cree phonotactics. For a final. Another oddity is that this layout sometimes opts to a full explanation, see Wolfart (1996). produce look-alike finals, instead of the characters specifi-
89 cally intended for Western Cree. Additionally, the keyboard has no key for the full-stop used in Plains Cree syllabics (namely: ᙮). 4.2 Other Layouts The Gboard application (van Esch et al., 2019) has an Android-only Cree syllabics layout. The layout is “pan- Cree”, aiming to cover all Cree dialects. Covering all di- alects in one layout has the effect that, out of 48 syllabics keys immediately available to type on the primary page, 18 keys (37.5%) are completely unused in Plains Cree syllabics Figure 3: The superposition of the original right-handed orthography. Given that the Gboard layout is Android-only, grid and the mirrored left-handed grid. Lighter areas are we will not discuss it further. To our knowledge, there are a worse places to assign keys, and darker areas are better few other smartphone keyboards for Cree syllabics (Houle, places to assign keys. 2018; Moshagen et al., 2016), however, we were unable to install them as the system-wide keyboard on our test device. As such, they will also not be discussed any further. indicated which 7mm targets are good places to assign keys, and which targets to avoid. 5. Design and Development of the Keyboard Now that we know which areas of the layout are easier and less error-prone to touch, and which areas are more error After studying the strengths and weaknesses of the layout prone, and more difficult to touch, we can begin placing presented in Section 4, we sought to make our own syllabics keys on the layout. layout for smartphones with the following goals: We decided, that in order to support two-thumbed typing, 1. All syllabics must be accessible on the primary page. we would create a “build-a-syllable” keyboard, where CV 2. Only keys used in Plains Cree should be on the pri- syllabics would be assembled by typing a consonant final mary page. first, then followed by the appropriate vowel. We placed 3. It must facilitate efficient two-thumbed typing. consonants and vowels on opposite side of the keyboard. 4. Its Unicode output must produce pre-composed syl- This way, when typing CV syllable sequences, the typist labic code points. alternates between thumbs, allowing simultaneous articu- lation of their thumbs. For example, while the left thumb With these goals in mind, we present our method for de- is finishing the articulation of typing a consonant, the right veloping this layout, which we will henceforth refer to as thumb is starting the articulation of typing a vowel. This fol- the Keyman layout. The Keyman layout is a “build-a- lows the design recommendations set by Norman and Fisher syllable” dynamic layout, designed specifically for use with (1982), who state that a keyboard should equalize the load the Plains Cree language. of both hands, maximizing sequences where keys are typed 5.1 Creating an Ergonomic, Two-handed Layout by alternating hands. Since Park and Han (2010) only collected data for right- Upon using the FirstVoices layout, the authors of this pa- handed touches, we made the assumption that this figure per felt that, in order to facilitate fast and accurate typing, can be mirrored horizontally to account for left-handed us- this layout could do better to improve the placement of the age. We then determined the width of a grid of 7mm wide syllabic characters. keys, with a 1mm gap between keys that would comfort- As such, we sought research in the ergonomics of thumbed- ably fit the width of a contemporary smartphone. Using a input on smartphones. We focused on two-thumbed input, smartphone with a screen width of 68mm, we calculated as most smartphone keyboard layouts position the virtual 8 × 7mm + 7 × 1mm = 63mm. Thus we determined that keyboard at the very bottom of the screen, where one would the grid should be 8 keys wide. We then superimposed both hold the phone with either one or two hands, using one’s the original right-handed grid, and the mirrored left-handed thumbs to tap on the keyboard’s keys. grid, assuming that keys in the overlapping portion of the Park and Han (2010) discuss the design and placement of grid will take the median of measures from both the left- touchscreen targets for one-handed interaction on a mobile handed and right-handed button placements (Figure 3). phone. The authors tested the effects of 5mm, 7mm, and 10mm targets on a 5 × 5 grid on a smartphone display. 5.2 Placing Keys Based on Corpus Statistics They measured study participants’ first transition time— We wanted to place frequent keys in the darker areas in how long it takes to go from a neutral position to any tar- Figure 3, and rare keys in lighter areas. To get an idea of get on the screen; the task completion time—how long it key frequency, we counted unigram and bigram frequency takes to press the correct target on the screen; the number in the Ahenakew-Wolfart corpus of Cree text (Arppe et of errors made; and the pressing convenience—the partici- al., 2019).2 This corpus is composed of a number of inter- pants’ subjective opinion of how easy it is to press a target. views and monologues recorded by Freda Ahenakew and H. We have calculated the median of all metrics measured in C. Wolfart and includes roughly 73,000 Cree word tokens. Park and Han (2010). We opted to use the 7mm grid, as it strikes a good balance between allowing for a great number 2The full results are available: https://gist.github.com/ of keys, and it is not as error-prone as smaller layouts. This eddieantonio/1b0f25f1c6d78e6dfb611f490a0822c7
90 5.3 Development with Keyman Developer ᕽ ᒼ ᐣ ᕀ ᐤ ᐃ ᐄ ᐆ As its name suggests, we made the Keyman layout using ᓬ ᑊ ᐠ ᐢ ᐋ ᐊ ᐅ ᕒ the Keyman Developer software (2020). To create a “build-a-syllable” dynamic layout—where some ᐨ ᐟ ᐁ ᐦ keys change based on the consonant just typed—we im- plemented the keyboard using several layers. “Layers” is ᙮ the term that Keyman Developer refers to as pages in the rest of this paper; however, while a page in a layout such Figure 4: The grid of our placements with syllabics keys as FirstVoices consists of completely different keys, in the filled in. Lighter areas indicate harder-to-type keys; darker Keyman layout, we employed layers to implement varia- areas indicate easier-to-type keys. tions of the primary page. We created Keyman layers for the primary page (the default layer) and created layers for all possible CV, and CwV con- Most texts in the corpus are some sort of narrative being told sonant “shapes”. All of these layers were identical, except to the interviewer. The corpus is written entirely in SRO, that the keys for the vowels were changed to reflect the pre- but since the mapping between SRO and syllabics is (for viously typed consonant. For example, starting from the the most part) straightforward, SRO letter frequencies are default layer (Figure 5a), typing the ᐠ (/k/) final switches to informative when creating a “build-a-syllable” keyboard. the kV layer (Figure 5b), which swaps all of the vowel syl- First, we placed generic keys, such as the spacebar, the re- labics with all possible kV syllabics, namely, ᑫ, ᑭ, ᑯ, ᑲ, turn key, the full-stop key (᙮) and the backspace key in ᑮ, ᑰ, ᑳ. Pressing any of these keys replaces the ᐠ with the areas that are frequent among QWERTY smartphone key- indicated kV syllabic. If instead, one presses ᐤ (/w/), the board layouts. Then, we placed the most frequent conso- keyboard switches to the kwV layer (Figure 5c), swapping nant, ᐠ (/k/) in the dark area on the left-hand side; then, we all the vowels once again with all kwV syllabics—ᑵ, ᑷ, ᑻ, placed the most frequent vowel, ᐊ (/ɐ/), in the right-hand ᑿ, ᑹ, ᑽ, ᒁ. Pressing a kwV syllabic replaces the ᐠᐤ with area of the grid. We continued this by placing higher fre- the appropriate syllabic. In this way, a syllabic is “built” by quency keys in the darker areas of Figure 3, and then placing first typing a final, then optionally typing ᐤ, and finally its lower-frequency letters in lighter areas. Consonants were corresponding vowel is selected. placed primarily on the left-hand side, while all 7 vowels Layer switching was accomplished using Keyman’s next- were placed on the right-hand side. Most of the placement layer directive for keys. In addition, each non-default layer of the keys were based on the corpus statistics, however visually highlights the keys that differ from the default some keys were placed for aesthetic value. For example, layer, as well as highlighting which consonant keys have the ᓬ (/l/) and ᕒ (/r/) were placed opposite of each other, been pressed (Figures 5b, 5c). since both are only used for loanwords from English and In total, the keyboard contains 19 layers: the default layer, French. The
91 (a) Default layer (b) kV layer (c) kwV layer
Figure 5: Switching layers in the Keyman layout. (Note: images have been scaled down).
(Section 6.2.2). The remainder of this section describes the wide, and 7.9mm thick. It has a diagonal screen size of 6.2 demographics of participants and the methodologies of the inches (157mm). The typing experiment was presented us- quantitative and qualitative analyses. ing an ad hoc web application, running in standalone mode. We have posted the source code of this application online.7 6.1 Participants The following methodology was repeated twice—once for For the purposes of this case study, we recruited three par- the FirstVoices layout, and then again for the Keyman lay- ticipants from the Edmonton, Alberta, Canada area. Partic- out. First, one of the two keyboard layouts was selected by ipants were recruited through open calls at the University the study facilitator. The study facilitator then left the room, of Alberta and targeted recruitment of Plains Cree speak- giving participants a chance to learn and practice using the ers with syllabics literacy. Participants ranged from 22–52 layout, privately, and at their own pace. We intentionally years of age (with a mean age of 32.67). Females com- did not teach participants how to use the selected layout; posed two of the three participants, and all three participants instead, participants had to teach themselves how to use it. identified as First Nations or Métis. Two of the three par- We presented 10 hand-picked words as prompts on a sin- ticipants completed all parts of the study. Every participant gle screen. Participants were asked to type the prompts in had an extensive background in typing on smartphones, reg- a text box positioned directly underneath it. These prompts ularly communicated in Plains Cree, and used syllabics on could be completed in any order. It was not mandatory for a smartphone or computer at least multiple times a week. participants to type these prompts accurately; it was only Given that testing focused on reproducing stimuli, fluency mandatory that each prompt was attempted. No data were in Plains Cree was not a requirement, but familiarity with collected during this period. syllabics and smartphones was. Figure 6 provides basic de- After the practice period, the participants asked the study fa- mographic information for participants and their relation- cilitator back into the room. Then, participants were given ship to both Cree and syllabics on a 5-point Likert scale a chance to ask the facilitator how to type specific charac- measuring how much they agreed with each statement. ter sequences. The study facilitator would instruct in a re- Although no participants had made use of our Keyman lay- hearsed and structured form, for any question asked by the out previous to this study, Participants A and C both had participant. Once satisfied, participants moved on to the prior experience using the FirstVoices layout. Participant B typing task. The facilitator instructed participants to type had no prior experience with either of the layouts studied. each prompt as quickly and as accurately as possible. The Although participants would have ideally had no knowl- facilitator left the room again to allow the participant to edge of either layout tested, considering that there are only a complete the typing task in private. We presented 30 sen- few options available for typing syllabics on a smartphone, tences of Plains Cree syllabics to participants for each lay- prior experience was unavoidable. Worth noting is that, de- out. We used a different set of 30 sentences for each layout, spite being able to read and write syllabics, Participant B resulting in 60 different sentences presented in the study. expressed discomfort in doing so. Before seeing each stimulus, participants were allowed to take a break, if desired. The sentences used as stimuli were 6.2 Methodology first prompted without a text box for 30 seconds to prime We invited each participant to a 90 minute study, conducted participants. This was to encourage participants to fully at the University of Alberta. The 90 minute study comprised read the stimulus. After the 30 second priming period, a text of a typing study to collect quantitative data regarding the box would appear, allowing participants to type the prompt. speed and accuracy of each keyboard layout, followed by a We manually constructed stimuli sentences to test the most short questionnaire to collect qualitative data. common syllabics observed in the Ahenakew-Wolfart cor- pus (Arppe et al., 2019). This was chosen over directly 6.2.1 Quantitative Methodology pulling sentences from the mentioned corpus because the To collect quantitative data regarding the speed and accu- conversational nature of the sentences required excessive racy of using both keyboard layouts, we had participants editing. Furthermore, many of the “clauses” in the corpus take part in a typing experiment. The typing experiment were defined by speech pauses, and include multiple clauses was facilitated by one of the authors. that made little sense without the requisite context. In order Participants were given a Samsung A10 smartphone run- to capture behaviour as naturally as possible, participants ning the Android 9.0 operating system for the duration of the typing test. The Samsung A10 is 155.6mm tall, 75.6mm 7https://github.com/eddieantonio/typing-test.
92 Participant Writing in Cree is important to me. ●●● ●●● It is important for me to write Cree on my phone. ● A I speak the Cree language ●● ● ● I am comfortable writing Cree in syllabics. ● ● ● B I am comfortable reading Cree in syllabics. ● ● ● ● Statement ● ● I am comfortable having conversations in Cree ● C SD D N A SA Agreement
Figure 6: Likert scale responses for participant demographics. Responses from left-to-right are strongly disagree, disagree, neutral, agree, and strongly agree. were not presented with individual words, but grammati- scribed for participant demographics). These ratings were cal clauses or sentences. Sentences were constructed to be composed of the following seven statements: fully grammatical clauses that could be easily understood, 1. I would use this keyboard again to type Cree syllabics. even if incomplete (e.g. the English subordinate clause: if 2. I would recommend this keyboard to my friends, fam- I ever see him again…). There are 95 individual syllabics ily, and/or students. used in this corpus. Because constructing a set that covers 3. Overall, I like using this keyboard. all 95 syllabics would require a large number of sentences, only those characters in the 90th percentile by frequency 4. I can type quickly with this keyboard. were tested. The remaining syllabics occurred fewer than 5. This keyboard was easy to use. 20 times in a corpus of over 200,000 characters. The num- 6. I can type on this keyboard without making mistakes. ber of sentences presented was capped at 30 per layout to 7. I could find each syllabic easily on this keyboard. maximize the number of syllabics tested without overly tax- Participants were also asked to give any feedback they had ing participants, as participants took roughly 80 minutes to in a free-form response for each layout. complete two sets of 30 sentences (60 sentences in total). The results of both the qualitative and quantitative assess- Data was collected starting when the participant tapped on ments are presented and described in the rest of this paper. the text box, issuing a focus event. Upon focusing the text box, the keyboard would pop-up, allowing the partic- ipant to type. We collected timing information for each 7. Results key tapped by the participant, including the backspace key. This section details the quantitative results (i.e., accuracies Keystroke events were collected in JavaScript by register- and averages) and qualitative results (i.e., questionnaire re- ing an event handler on the HTML
93 FirstVoices Keyman mend the FirstVoices layout to friends and family, and that they would use the keyboard again. Participant C reported Participant A 98.88% 93.70% a strong agreement toward recommendation and regarding Participant B 93.70% N/A the ease to find keys on the keyboard (henceforth naviga- Participant C 96.92% 98.42% bility). Results for the Keyman layout were more varied. Mean 96.50% 96.06% Participant B strongly disagreed with all statements, sug- gesting a very negative impression of this layout. Partici- Table 2: Keyboard Accuracies pant C disagreed with all statements, except for describing ease of use, a statement with which they agreed. Participant A had mixed feelings: they disagreed that the Keyman lay- the Keyman layout errors were largely those of spacing: out was quick and navigable, were neutral as to how easy the Keyman layout’s novel feature— the thin, non-breaking the keyboard was to use and how likely they were to recom- space—was never used by Participant A and was frequently mend it, and agreed that they would use the keyboard again, missed by Participant C. Spacing errors alone made up 31 liked the keyboard, and that the keyboard was accurate. out of 43 (72%) of the Keyman layout’s errors. Since the In freeform feedback for the keyboards, one participant re- thin spaces are a feature of the Keyman layout alone, when ported issues with the FirstVoices spacebar being too small, we disregard these errors, the overall accuracy of the Key- and that the ᐦ key was too close to the backspace key, caus- man layout rose to 98.8%, two percent points higher than ing accidental deletions throughout the session. Participant the FirstVoices keyboard. A mentioned that they like the layout of the FirstVoices lay- out, and that once they realized each row contained syllabics ● 10.0 with the same vowel, it became easy to use. Regarding the Keyman layout, one participant described ● their issues as being that “This keyboard required too much ● thinking, slowing down the typing process. I liked the idea 7.5 ● ● about it and its simplicity, but after a while I started not to ● Keyboard ● ● like it just because it took too long.” Another participant FirstVoices suggested that our layout was too rooted in English ideol- 5.0 ogy, and that they did not think of syllabics of being proce- Keyman durally generated first by a consonant to determine shape, ● and then by a vowel to determine orientation. This partici- 2.5 ● pant further suggested adopting a keyboard layout reminis- cent of the “star chart” layout as described in Houle (2018). Although this layout is familiar to some, it is by no means learned by the majority of syllabics users. Time per Character per Sentence (s) Time per Character A B C Both layouts were criticized for being too small by multiple Participant participants. They reported poor accuracy due to not being able to read key labels. One participant noted that the ex- Figure 7: Average time-to-type per sentence (in seconds). clamation mark was unavailable on both layouts. In fact, this symbol was available on the Keyman layout, but re- In addition to accuracy, we calculated the average time it quired a long press to be accessed and so went unnoticed by took to type each character for each stimuli sentence. Be- this participant. Finally, one participant reported generally cause the number of characters varied between stimuli sen- “having issues” with both keyboards and wanting a solu- tences, we opted to divided the total time it took to type each tion for mobile typing of Plains Cree syllabics; they made it sentence by the number of characters in that sentence. This clear, however, that the Keyman layout was not the solution measure, time-to-type, was used to evaluate how fast a par- they were looking for. ticipant was able to type on each layout. Figure 7 plots the average time to type a character per stimuli sentence. 8. Discussion The results described above expressed a general preference 7.2 Qualitative Results for the FirstVoices layout. On nearly every dimension, the Participants as a whole preferred the FirstVoices layout. FirstVoices layout was preferred over the Keyman layout, Figure 8 plots participant responses to each of the questions regardless of which participant was doing the ratings. It detailed in Section 6.2.2. The labels on the y-axes represent is worth noting, however, that Participants A and C report abbreviated forms of the questions from Section 6.2.2. previous familiarity with the FirstVoiceslayout. Given their All three participants generally agreed with all statements extremely limited exposure to the Keyman layout, the lack for the FirstVoices layout, indicating positive attitudes to- of confidence in using it is understandable. Accuracies were ward the keyboard. The only negative ratings were from comparable between keyboards, though the FirstVoices’ Participants A and C who disagreed that the keyboard al- was slightly higher. However, most of the Keyman lay- lowed for accurate typing. Participant A also reported a out’s errors were due to not using the non-breaking space, neutral response toward the keyboard being quick to type a feature not found in the FirstVoices layout. When ignor- on. Notably, all participants agreed that they would recom- ing these errors, the Keyman layout proved more accurate
94 FirstVoices Participant Would use again ●●● Would recommend ●● ● Like ● ● ● A Is quick ● ●● Is easy ●● ● ●● ● B Is accurate ●● ● Statement Can navigate ● C SD D N A SA Agreement Keyman Participant Would use again ● ● ● Would recommend ● ● ● Like ● ● ● ● A Is quick ● ●● Is easy ● ● ● ● B Is accurate ● ● ● ● ● ● Statement Can navigate ● C SD D N A SA Agreement Figure 8: Responses to usage questionnaire about the layouts. than FirstVoices keyboard. Further, the FirstVoices layout the Keyman layout, their opinions could change. We in- was observed to have a broader ranger of errors, though this tentionally avoided giving participants explicit training, but may be due to the fact that it was the only keyboard used by the Keyman layout proved unintuitive; however, it would Participant B, who exhibited a much higher error rate than be interesting to know the opinions of typists who have the other two participants. The quantitative data suggest gained significant experience on the layout, rather than the that despite being a more accurate keyboard, the Keyman 90 minute session presented in this paper. Future evalua- layout was less pleasant to use. tion could require a longer timescale, allowing participants In terms of qualitative assessments, all participants pre- to learn and practice the layouts. ferred the FirstVoices layout. Although some amount of preference is expected for a familiar keyboard layout, par- 9. Conclusion ticipants showed slower typing speeds on the Keyman lay- out and commented on the amount of effort it took to type on This paper has described an attempt to build a mobile key- the Keyman layout. Although built from the ground up as a board for writing in Plains Cree syllabics. To address com- Cree-based keyboard, one that eschews European baggage, munity calls for a Cree worldview based syllabics keyboard, the Keyman layout supposed some amount of composition- we built a keyboard layout using the Keyman infrastructure. ally for the characters. Asking participants to first type a This keyboard required participants to first select the con- final version of a consonant and then presenting them with sonant of the syllabic they wished to type before presenting all syllabics using that consonant as an onset appeared to them with a set of syllabics using that consonant as an on- be unintuitive. Compared to the FirstVoices layout, which set. As there is no English on this keyboard, participants is a straightforward listing of all syllabics, participants had had to choose a consonant value based on the consonant’s a harder time navigating through the keyboard to type the final character. This keyboard was tested in a case study appropriate character. This in turn required a lot of thought. along side the existing FirstVoices layout for comparison. Considering that Plains Cree syllabic finals do not transpar- Although accuracy was higher on the Keyman layout, par- ently map to the associated shape of their onset counterparts ticipants unanimously preferred the FirstVoices layout. The (e.g. ᐠ and /k/ shares no similarities to the shape of syllabics Keyman layout was reviewed as hard to use, requiring much with /k/ as an onset, like ᑲ), it is possible that participants mental effort, and being unpleasant, and slow. The results struggled to quickly determine which final was needed for of this paper indicate that, while there is a desire for alter- a given syllabic shape. This disconnect may have signifi- natives to the solutions that currently exists, there is a high cantly contributed to the experience of using the keyboard. cost in introducing an unfamiliar typing system and asking people to adapt to it. In addition to mental effort, the Keyman layout required a lot of comfort and familiarity with syllabics. It is unsurpris- ing that Participant B, who was least comfortable in syllab- 10. Acknowledgements ics, was completely unwilling to use the keyboard. Thanks to Marilyn Shirt for prompting this project. Thanks Efforts to improve the ergonomics seemed not to factor to Marc Durdin, Joshua Horton, and Darcy Wong for putting much into participants reviews. Despite community mem- up with our constant bug reports. Thanks to Lorna Williams bers disliking the current options available for typing Plains and David L. Rowe for helping us publish the keyboard Cree—including the FirstVoiceslayout—participants strug- on Keyman’s website. Thanks to Anna Kazantseva and gled throughout the case study with the new Keyman lay- Roland Kuhn for their endless support. Thanks to the fol- out. It remains unclear as to whether or not this is an effect lowing people, in no particular order, for help with this pa- of having little experience with the keyboard. It is plau- per: Aidan Pine, Antti Arppe, Delaney Lothian, Rebecca sible that, once participants familiarized themselves with Knowles, and Katherine Schmirler. ᑭᓇᓈᐢᑯᒥᑎᓈᐣ᙮
95 11. Bibliographical References Ogg, A. (2018). Private communication, September. Arppe, A., Schmirler, K., Harrigan, A. G., and Wolven- Park, Y. S. and Han, S. H. (2010). Touch key design for grey, A. (2019). A morphosyntactically-tagged corpus one-handed thumb interaction with a mobile phone: Ef- for Plains Cree. In Maculay M. et al., editors, Papers of fects of touch key size and touch key location. Interna- the Forty-Ninth Algonquian Conference, East Lansing. tional journal of industrial ergonomics, 40(1):68–76. Michigan State University Press. Santos, E. A. (2019). OCR evaluation tools for the 21st Cassels, M. (2019). Indigenous languages in new media: century. In Proceedings of the Workshop on Computa- Opportunities and challenges for language revitalization. tional Methods for Endangered Languages, volume 1, Working Papers of the Linguistics Circle, 29(1):25–43. page 4. First Peoples’ Cultural Council. (2016). FirstVoices Stevenson, W. (1999). Calling Badger and the Symbols of apps. https://www.firstvoices.com/content/apps/, the Spirit Language: The Cree Origins of the Syllabic May. (Accessed on 02/03/2020). System. Oral History Forum, 19–20:19–24. Houle, J. R. (2018). ᒪᓯᓇᑕᐦᐃᑫᔨᐣ ᐊᒐᐦᑭᐯᐦᐃᑲᓇ; masi- The Truth and Reconciliation Commission of Canada. natahikeyin acahkipehikana; typing syllabics. Mas- (2015). What we have learned: Principles of truth and ter’s thesis, University of Alberta, Edmonton, Alberta, reconciliation. http://nctr.ca/assets/reports/Final% Canada, September. 20Reports/Principles_English_Web.pdf. Keegan, T. T., Mato, P., and Ruru, S. (2015). Using Twitter The Unicode Consortium, (2019a). The Unicode Standard, in an Indigenous language: An analysis of Te Reo Māoria Version 12.1.0, chapter 6, page 265. The Unicode Con- tweets. AlterNative: An International Journal of Indige- sortium, Mountain View, CA. http://www.unicode. nous Peoples, 11(1):59–75. org/versions/Unicode12.1.0/. Keyman. (2020). Keyman | type to the world in your lan- The Unicode Consortium. (2019b). Unified Canadian guage. https://keyman.com/, January. (Accessed on Aboriginal Syllabics range: 1400–167F. In The Uni- 02/03/2020). code Standard, Version 12.1.0. The Unicode Consortium, McIlroy, R. and Kyöstilä, S. (2018). Clamp Mountain View, CA. https://unicode.org/charts/ performance.now() to 100us. https:// PDF/U1400.pdf. chromium.googlesource.com/chromium/src/+/ van Esch, D., Sarbar, E., Lucassen, T., O’Brien, J., Breiner, a77687fd89adc1bc2ce91921456e0b9b59388120% T., Prasad, M., Crew, E., Nguyen, C., and Beaufays, F. 5E%21/, January. (Accessed on 02/06/2020). (2019). Writing across the world’s languages: Deep in- Moshagen, S. N., Siewertsen, E., and Gaup, B. (2016). ternationalization for Gboard, the Google keyboard. giellalt/keyboard-crk: Plains Cree keyboard layout. Wolfart, H. C. (1996). Sketch of Cree, an Algonquian lan- https://github.com/giellalt/keyboard-crk, May. guage. In Ives Goddard, editor, In Handbook of North (Accessed on 02/10/2020). American Indians, volume 17, pages 390–439. Smithso- Norman, D. A. and Fisher, D. (1982). Why alphabetic key- nia Institution Washington, DC, USA. boards are not easy to use: Keyboard layout doesn’t much Wolvengrey, A. (2018). Private communication, August. matter. Human Factors, 24(5):509–519.
96